On Mon, Apr 23, 2012 at 10:48:34AM -0400, Ted Ts'o wrote: > On Mon, Apr 23, 2012 at 03:22:52PM +0300, Touko Korpela wrote: > > > Finally, the only file system where someone is likely to be creating > > > that is this small in this day and age is the /boot filesystem --- and > > > there, even if the drive using 512-byte emulation, performance isn't > > > an issue since no one is executing out of /boot, or even modifying it > > > very often. > > > > Yes, /boot is my primary worry. > > Yes, and the performance hit only really happens when you need to do a > read-modify-write cycle. And /boot doesn't get modified very often > --- only when you update a kernel, and even when you do, it's mostly > large files which will generally be contiguous. So the performance > hit is barely measurable. > > > A thing to consider is that "dumb" SSDs and > > USB sticks/memory cards are more common than "smart" SSDs. > > (a) dumb SSD's still had to work well on Windows XP, and hardware > designs are conservative, so it's not clear to me this is really a > huge issue. And (b) USB sticks/memory cards are again primarily used > for file interchange, where performance is not a big thing. (It's > really random 4k writes where the read-modify-write cycles really > hurt. For big files, we where we issue a large contiguous write > transaction, you only do the read-modify-write cycle for the first and > last 4k block and that's not a big deal.
USB sticks and other flash media are optimised for FAT (using big blocks). Most of my information comes from LWN article "Optimizing Linux with cheap flash drives" https://lwn.net/Articles/428584/ I don't know if linear reads during boot are affected but other usage is much slower when reads/writes aren't aligned. Fdisk has already fixed its default partition layout. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
